Answers

How to see which sources ChatGPT cites

Most brand-visibility work focuses on whether you're mentioned. The bigger lever is usually further upstream: which pages ChatGPT is actually pulling from.

When ChatGPT uses web browsing, its answer includes clickable source links — open a category or comparison prompt, look for the small citation markers or the linked domains beside the answer, and record every URL across a prompt set of 20-30 queries. Group them by domain to see which handful of sites ChatGPT actually treats as authoritative for your category.

  • Citations only appear when ChatGPT's browsing tool is triggered — not every response includes them.
  • A small number of domains usually account for most citations in any given category.
  • Getting listed on a cited third-party page often moves visibility faster than publishing new content yourself.
  • Citation patterns differ by query type — definitional vs comparison vs "best of" queries cite differently.
  • This is the single highest-leverage audit for figuring out where to focus content and PR effort.

Why the citation list matters more than the mention

Most people trying to improve AI visibility start by asking: "does ChatGPT mention us?" That's the wrong first question. The better one is: "which pages is ChatGPT actually reading to write this answer?"

When ChatGPT's browsing tool is active — which happens automatically for queries that benefit from current information, like comparisons, pricing, or "best X" questions — the answer is built from a small set of retrieved pages, and those pages are shown as clickable citations. That list is a direct window into which sources the model currently trusts for that topic.

If your brand isn't showing up in answers, the citation list usually tells you why faster than any amount of guessing: either none of the cited sources mention you, or they do and the model just isn't surfacing that detail yet.

Not every ChatGPT response triggers browsing. Definitional or general-knowledge questions often answer from training data alone, with no citations at all — that's expected, not a bug in your test.

How to run the citation audit

  1. 1

    Use prompts that are likely to trigger browsing

    Comparison queries, "best X for Y in 2026," pricing questions, and anything time-sensitive are far more likely to trigger a live search than a general definitional question. Build your prompt set around these patterns specifically for this audit.

  2. 2

    Open each cited link, not just the domain

    Note the exact URL, not just the root domain — a specific blog post behaves differently from that same site's homepage. The exact page tells you what kind of content is getting pulled.

  3. 3

    Tally citations by domain across the full prompt set

    Run 20-30 prompts, collect every citation, then group by domain. In most categories, three to six domains account for the majority of citations — that concentration is the useful part of this exercise.

  4. 4

    Check whether you're mentioned on each frequently-cited page

    For the domains that show up repeatedly, search their site for your brand name. If you're absent from a page ChatGPT is already citing for your category, that's a faster fix than trying to get a brand-new page ranked from zero.

The kinds of sources that tend to get cited

Patterns vary by category, but these show up repeatedly across B2B software audits.

  • Review aggregators

    G2, Capterra, and category-specific review sites are cited heavily for "best X" and comparison queries.

  • Independent comparison content

    Third-party blog posts and roundups that already compare multiple tools by name.

  • Community discussion

    Reddit and similar forums are cited more than most brands expect, especially for candid "is X worth it" queries.

  • Industry publications

    Trade press and analyst-style writeups, particularly for category-defining or "what is X" queries.

  • Your own site

    Product and comparison pages on your own domain, when they're specific enough to be directly useful.

What to actually do with the list

Split the citation list into two workstreams, because they need different owners and move on different timelines.

Owned pages that get cited are a signal to double down — the model already trusts that content shape and depth, so extend the same treatment to adjacent topics on your own domain.

Third-party pages that get cited but don't mention you are an outreach and relationship task, not a content task. Getting added to an existing, already-trusted comparison page or review listing is frequently faster than trying to out-rank it with a brand-new page.

Frequently asked questions

Why doesn't my ChatGPT response show any citations?
Citations only appear when ChatGPT's browsing tool activates, which happens for time-sensitive, comparative, or current-information queries. General or definitional questions are often answered from the model's training data with no live search and no citations — that's expected behavior, not a sign your test failed.
Is the citation list the same as a Google search results page?
No. ChatGPT's browsing tool runs its own retrieval and the model decides which retrieved pages to actually reference in the final answer, which is a smaller and differently-weighted set than a full search results page. A page ranking well in Google isn't automatically cited by ChatGPT.
How many domains typically show up in a citation audit?
In most B2B categories, a relatively small set — often three to eight domains — accounts for the majority of citations across a 20-30 prompt audit. That concentration is exactly what makes the audit useful: it turns an open-ended problem into a short, specific list of pages to influence.
Should I track this once or on an ongoing basis?
Ongoing. Which sources get cited shifts as new content gets published and re-crawled, and as the model itself updates. A one-time audit tells you where things stand today; a repeated one tells you whether your outreach and content efforts are actually changing the citation mix.

See your full citation profile automatically

EvidentlyAEO tracks every source ChatGPT, Gemini, Perplexity, and Claude cite for your category — ranked by how often they actually influence answers.

Related answers